You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Oct 22, 2023. It is now read-only.
When doing aspect ratio bucketing, there are several predefined bucket groups, defined in dataloaders_util.py. For example, if I choose the 640 bucket group, I will get buckets like 640x640, 704x576, 768x512, 896x448, etc.
For my experiments, I would like to use a bucket that ALWAYS has 512 as one of dimensions. So the buckets would be like 512x512, 512x576, 512x640, 512x768, etc. I realize that the different buckets would have different memory requirements, but there are some reasons why this might work better.
To solve this myself, I tried to just edit the buckets in the dataloaders_util.py myself, but I get an immediate error after doing so, and I'm not familiar enough with the codebase to figure out the reason. Please consider making a way to add custom bucket groups.
The text was updated successfully, but these errors were encountered:
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.
When doing aspect ratio bucketing, there are several predefined bucket groups, defined in dataloaders_util.py. For example, if I choose the 640 bucket group, I will get buckets like 640x640, 704x576, 768x512, 896x448, etc.
For my experiments, I would like to use a bucket that ALWAYS has 512 as one of dimensions. So the buckets would be like 512x512, 512x576, 512x640, 512x768, etc. I realize that the different buckets would have different memory requirements, but there are some reasons why this might work better.
To solve this myself, I tried to just edit the buckets in the dataloaders_util.py myself, but I get an immediate error after doing so, and I'm not familiar enough with the codebase to figure out the reason. Please consider making a way to add custom bucket groups.
The text was updated successfully, but these errors were encountered: